LabNotes2

advertisement
Csci2412 Lab work
Laboratory Session 2 – further introduction to the
Matlab package.
Learning outcomes
Know the connection between matrix computations and neural network design in matlab.
Know the basic constructs of the matlab language.
The language
Matlab is a programming language. It allows assignment of variables.
Try entering the following in the command window (hit return at the end of each line)
A=5
B=’toodle do’
C=67
A+B
As you can see matlab is not strongly typed. Hazard a guess at what is happening?
Ending a command with ; stops the echo.
A=8;
You can view the workspace and edit variable values directly – change A, B etc via the
workspace view now. [The workspace is accessed via a tab in the upper left hand
window. Ask if you can’t find it]. Double click a variable to see its contents – if you want
to change a value.
There are the usual kinds of control structures
Enter
for i=1:3 A=A+1 ,end
and see what happens.
matlab also has if … else statements, while statements etc. See the tutorials in help under
flow control for more info.
Matlab help
Matlab help will give you information about any command – go to the command window
and just type help followed by the command. Try
Csci2412 Lab work
help help
If you want to keep track of what you are doing try the diary command (use the help to
find out about it but do it after this lab);
If you want to keep all your working you can save the workspace as a .mat file.
You can also use help to find out about "for" statements etc.
Calculations
Re-assign B to a number and try some calculations
B=23
D=A*B+C
Entering and calculating with matrices
Matlab is suited to working with arrays of numbers called matrices.
We can enter a matrix as follows and Matlab will respond.
a=[1 2; 1 2]
Now enter
b= [1 3;1 4]
There are add and multiply operations on matrices which should have been mentioned in
a lecture. Look at the output from the following.
a+b
a * b
You can probably guess how add works (but probably not guess about *).
Lets have a new matrix c
c=[1 2 3; 3 4 5]
We can't always multiply matrices if they are incompatible in size. Lets try to multiply
c*a
Csci2412 Lab work
c * a
We can get a * c though
a*c
Remember the compatibility rules for matrices from the lecture. We will not be
multiplying matrices by hand but will need to understand the link between matrices and
neural nets – which we explore further below. Understanding the compatibility rules for *
will be important.
You can select rows or columns within a matrix easily in matlab
c(1,:)
gives the first row and
c(:,2)
gives the second column.
This ability to pick rows and columns out is very useful. You can also pick out multiple
columns and rows as we will see in the next lab.
If you want to transpose your columns and rows (quite a common thing to do) you can do
it as follows
c=c’
Importing and exporting data and saving files:
Use the file menu to import the workspace that you saved last time. This gives access to
all the data we had last time. Notice that you will lose any data you had before the import.
The matrix p is not a big matrix – but it might be and we might want to save it to a file
save ‘myData.txt’ p -ascii
Look at this file to see what you have got. This is a useful option if you want to do some
processing with matlab but use the output with another program. It can be useful to
manipulate data etc. in excel and then process in matlab before passing back to excel (or
the other way around).
Using save to keep part of the workspace can be useful as well
save data network1
will keep the network in the file data.mat.
We can load files as well – be careful with the quote marks. They are essential.
z=load('myData.txt')
Csci2412 Lab work
Plotting graphs
matlab is useful for plotting pictures – we shall use it to plot some straight lines.
First create some points at uniform intervals,
x=1:0.1:4;
then calculate the corresponding y values.
y=3*x+4
then plot them
plot(x,y)
[If you want to know the effect of the x=1:0.1:4 command look at the output x in the
workspace window.]
If we want to do something repeatedly we create an m-file. An m-file is a script – it can
refer to variables in the workspace if desired and it can behave like a function. Make sure
that you have a copy of showLine.m in your current directory and type
showLine(2,-2,x)
Look at the content of showLine.m to see what an m-file looks like. You can look at the
m-files which lie behind all of the matlab functions if you want.
The connection between neural nets and matrices (in matlab)
Your workspace should still contain network1 and the input output pairs held in p
(inputs) and t (target outputs) because you imported them from the last lab.
p(1) and t(1) are an input target pair (let’s remind ourselves what they look like)
p(1)
t(1)
Open the NNtool and import network1 into the tool and view it via the view button on the
network manager (highlight network1 first). This picture contains lots of information.
Identify the places on the picture where the following information is provided
The network expects a 2 place input vector
The network has a layer which consists of 1 neuron which uses the hardlim transfer
function.
We can access the internal pieces of this network
w=network1.IW{1,1}
b=network1.b{1}
Now we can compute with the matrix values that are there – but first notice how many
rows there are in w and compare to the number of neurons in the perceptron layer.
Compare the number of columns in w with the input size.
Check that
Csci2412 Lab work
hardlim(w*[1;2] +b)
sim(network1,[1;2])
have the same output as each other.
Let us get some more data into our workspace and create a new network to make sure we
have the complete idea.
We will choose input vectors of size 3
p= [1 2 3 4 5;2 3 4 5 6; -4 -4 -5 -10 7]
and outputs of size 2 (so we need a perceptron layer with two perceptrons when we create
the network)
t= [0 1 1 0 1;1 0 0 1 0]
Delete all the old information in the network manager and import the new p and t values.
Create and train a neural network (network2) which learns this input - output pattern
(remember to create a perceptron network and that you need 2 neurons)
View your network and look at the matrix values associated with it. Again check the
relationship between the number of rows in w and the number of neurons in the network
and the number of columns in w and the size of the input vector.
w=network1.IW{1,1}
b=network1.b{1}
Now compare
hardlim(w*p(:,1) +b)
with
sim(network1,p(:,1))
Check that network1 computes the same value on p(:,x) as hardlim(w*p(:,x) +b)) does for
each column of p.
Csci2412 Lab work
Useful neural net functions
function name
adapt
adaptParam.passes
hardlim
adaptwb
init
initzero
learnp
learnParam.lr
learnpn
logsig
newff
learnwh
newlin
newlind
newp
newrb
purelin
sim
train
trainParam.epochs
action/use
allows a nn to adapt
the number of passes for which the network adapts
hardlimit function
adapts causing weight and bias changes according to learning parameters
initialises the network according to network spec
initialises with zero weights and biases
perceptron learning function
the learning rate
normalised perceptron learning function
log sigmoid transfer function
creates new feed forward network
widrow-hoff learning function
creates a new linear network or layer
designs anew linear layer
creates a new perceptron
creates new radial basis network
linear transfer function
simulates a network
trains a neural network
the number of epochs for which to train
To get help for any of these functions type
>>help "name"
at the matlab command prompt or possibly
>>help network/"name"
if the first option doesn't provide information.
Download